Clasificación de galaxias utilizando procesamiento digital de imágenes y redes neuronales artificiales

Abstract

Palabras clave: análisis de datos, predicción y clasificación, aprendizaje de máquinas, red neuronal artificial convolucional, analisis de componentes principales, ACP, RNA, inteligencia artificial, reconocimiento de patrones, clasificación de galaxias, morfología de galaxias.Abstract. The study of the formation and evolution of galaxies requires the measurement of their morphological parameters. Traditionally, 295 morphological analyses have been performed through the extraction and selection of features, or through visual inspection by experts in a highburden process which is almost impossible to perform in massive image collections. Although there have been several attempts to build automated classification systems, these do not possess the required precision level. In this work we developed a convolutional artificial neural network and trained it with the massive database of the Galaxy Zoo project. This neural network can be applied to the automatic classification of images of galaxies according to their morphology. To increase the precision in the classification of the neural network, the training images are preprocessed using a principal component analysis approach. By reducing the work burden of experts and by not depending on the inexpert manual interpretation of images, this scope could be fundamental for the analysis and classification of images coming from even wider surveys which are currently under development.

Conceptual low-cost on-board high performance computing in CubeSat nanosatellites for pattern recognition in Earth's remote sensing

“…It has been noted that a hybrid ANN/PCA algorithm constitutes a good framework to the classification problem. In [4], it has been successfully applied to solve the problem of massive classification of galaxy structures in large image data-sets. In this work we reproduce such paradigm with a convolutional ANN and PCA to perform real-time object recognition and classification in a set of high-resolution images, which would benefit a CubeSat space mission to analyse on-board large sets of images taken with high-resolution payloads.…”